A Shop That Knows You Better Than You Know Yourself? AI in Retail

In spring 2014, researchers at Northeastern University ran an experiment that was simple in concept: hundreds of real users and a set of synthetic accounts visited 16 commercial websites, and what each of them saw was compared. The researchers found that some sites showed different prices, or a different ordering of results, depending on the device, membership status, and browsing and purchase history, and that in some cases the differences ran to hundreds of dollars. But they also found that most of the experiments revealed no price discrimination or steering. That mixed picture is the most honest way into the subject: a real capability, and an extent that remains unsettled.

From the trader's instinct to the algorithm's forecast

Trade has never been far from prediction; a skilled merchant reads his customer and puts the right item within reach. What is new is that machines do this at a scale no individual's instinct can reach: forecasting demand for an item in a particular branch, suggesting to each visitor what suits their history, and adjusting what is ordered from suppliers.

McKinsey (2023) estimates that generative AI could add between $400 billion and $660 billion a year to retail and consumer packaged goods, equivalent to 1.2 to 2.0 per cent of their revenues. Fairness requires reading the figure as an estimate of potential, measured by the institute's models, not as profit already earned.

When the price becomes an opinion about you

The question goes beyond recommendations to pricing. In January 2025 the US Federal Trade Commission published preliminary findings from a study based on documents from six intermediary companies. They showed that software available to retailers can draw on a wide range of consumer data: location, demographics, browsing and purchase history, mouse movements, and items left in the shopping basket.

Yet the Commission itself gave only hypothetical examples. It did not confirm particular cases of individualised prices imposed on buyers, and it described the study as still under way when it published. The accurate statement is therefore that the capability is documented, while the extent of its use needs more evidence.

Not every price difference is an injustice; a discount for a member or a loyal customer is legitimate. The problem lies in concealment and deception, and in the jurisprudence of transactions there is a firm principle that the seller must make things clear and that fraud is forbidden. This is why transparency is a requirement, not an ornament: the buyer should know when a paid recommendation is being shown to them, and on what basis prices change.

The bright side: less food in the bin

Good demand forecasting has a benefit that is uncontroversial in principle: a shop that orders in line with what it sells throws away less. The scale of the problem is large; the UN Environment Programme's report (2024) estimated that the world wasted 1.05 billion tonnes of food in 2022, about 19 per cent of the food available to consumers.

The distribution of that waste tempers the enthusiasm, however. Households account for 631 million tonnes, food services for 290 million, and retail for only 131 million. A shop's intelligence therefore addresses the smallest slice; the largest lies in our own habits of buying, cooking and storing. And wastefulness is something every Muslim should reconsider, grateful for the blessing and careful to preserve it.

Data and competition

Every precise recommendation rests on the buyer's data, and this is where regulation comes in. In the Kingdom, the Personal Data Protection Law was issued by Royal Decree M/19 (September 2021). It lets individuals learn how their data is processed, access it, correct it, and request its destruction when it is no longer needed, and it requires consent or an opt-out mechanism before data is used for marketing. For more on this, see the article "Privacy in the Age of AI".

As for the effect on competition, we offer a cautious inference: large platforms gather from transactions what a small shop cannot, so the gap between those who own data and those who rent tools to analyse it is likely to widen.

What to do

• For the shopper: compare prices from more than one device or browser before a large purchase, review privacy settings, and do not treat an insistent recommendation as innocent advice.
• For the shop: make personalisation a service to the customer, not an exploitation of their moment of weakness or haste, and state your pricing policy clearly.
• For regulators: monitor actual practice before building rulings upon it, and require disclosure where it is needed.

The question these tools bring to every humble seller is this: will he use what he knows about his customer to serve him, or to single him out? The algorithm does not answer that question; the one who owns it does.